[Mesa-dev] [v4 09/11] i965: Set render state for lossless compressed

[Pohjolainen, Topi]

On Thu, Apr 21, 2016 at 02:59:04PM +0300, Topi Pohjolainen wrote:
> v2: Add support for blorp and removed the support for meta
> 
> Signed-off-by: Topi Pohjolainen <topi.pohjolainen at intel.com>
> ---
>  src/mesa/drivers/dri/i965/brw_blorp_blit.cpp  | 3 +++
>  src/mesa/drivers/dri/i965/brw_blorp_clear.cpp | 6 ++++++
>  src/mesa/drivers/dri/i965/brw_draw.c          | 7 ++++++-
>  3 files changed, 15 insertions(+), 1 deletion(-)
> 
> diff --git a/src/mesa/drivers/dri/i965/brw_blorp_blit.cpp b/src/mesa/drivers/dri/i965/brw_blorp_blit.cpp
> index 74fe3c0..84f4ca5 100644
> --- a/src/mesa/drivers/dri/i965/brw_blorp_blit.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_blorp_blit.cpp
> @@ -107,6 +107,9 @@ brw_blorp_blit_miptrees(struct brw_context *brw,
>     brw_blorp_exec(brw, &params);
>  
>     intel_miptree_slice_set_needs_hiz_resolve(dst_mt, dst_level, dst_layer);
> +
> +   if (intel_miptree_is_lossless_compressed(brw, dst_mt))
> +      dst_mt->fast_clear_state = INTEL_FAST_CLEAR_STATE_UNRESOLVED;
>  }
>  
>  static int
> diff --git a/src/mesa/drivers/dri/i965/brw_blorp_clear.cpp b/src/mesa/drivers/dri/i965/brw_blorp_clear.cpp
> index b1da935..bf8d231 100644
> --- a/src/mesa/drivers/dri/i965/brw_blorp_clear.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_blorp_clear.cpp
> @@ -398,6 +398,12 @@ do_single_blorp_clear(struct brw_context *brw, struct gl_framebuffer *fb,
>         * redundant clears.
>         */
>        irb->mt->fast_clear_state = INTEL_FAST_CLEAR_STATE_CLEAR;
> +   } else if (intel_miptree_is_lossless_compressed(brw, irb->mt)) {
> +      /* Compressed buffers can be cleared also using normal rep-clear. In
> +       * such case they bahave such as if they were drawn using normal 3D
> +       * render pipeline, and we simply mark the mcs as dirty.
> +       */
> +      irb->mt->fast_clear_state = INTEL_FAST_CLEAR_STATE_UNRESOLVED;

We discussed in irc with Ben that this should be only hit with partial
clears. I added:

         assert(partial_clear);

I didn't see this trigger, and Ben, I didn't actually see assertion failures
in your test run either. The two deqp regressions:

piglit.spec.arb_pixel_buffer_object.texsubimage pbo.sklm64
piglit.spec.arb_pixel_buffer_object.texsubimage-unpack pbo.sklm64

are actually real rendering errors that I haven't seen before. We analysed
this a little with Ken and were able to make these pass with:

diff --git a/src/mesa/drivers/dri/i965/brw_blorp_clear.cpp b/src/mesa/drivers/dri/i965/brw_blorp_clear.cpp
index 7cfaae7..363b558 100644
--- a/src/mesa/drivers/dri/i965/brw_blorp_clear.cpp
+++ b/src/mesa/drivers/dri/i965/brw_blorp_clear.cpp
@@ -30,6 +30,7 @@
 #include "util/ralloc.h"
 
 #include "intel_fbo.h"
+#include "intel_reg.h"
 
 #include "brw_blorp.h"
 #include "brw_meta_util.h"
@@ -435,6 +436,9 @@ brw_blorp_resolve_color(struct brw_context *brw, struct intel_mipmap_tree *mt)
 
    brw_blorp_exec(brw, &params);
    mt->fast_clear_state = INTEL_FAST_CLEAR_STATE_RESOLVED;
+
+   brw_emit_pipe_control_flush(brw,
+                               PIPE_CONTROL_RENDER_TARGET_FLUSH);
 }
 
 } /* extern "C" */


In bspec there is:

"When performing a render target resolve, PIPE_CONTROL with end of pipe sync
must be delivered."